Topic: That where it is possible to esteem on Iskustvennomu to Intelligence
.. To be exact about it of now almost implementable part - deep networks. What input for dummies. To me the one-layer trained network is now clear is as a matter of fact NxM a matrix, it train fining (changing coefficients) responsible for wrong answers. How business from the multilayer is? I read that train each layer separately, but how? What is criterion of the wrong answer of the intermediate layer?